[Bayesian Optimization]Kriging Believer and Constant Liar

Both KB and CL are improvements based on the q-EI method. is an approximate parallel strategy. Please point out if there is something wrong!

Ordinary q-EI:

Remember the black box functiony(x)

Select q points in each round of iterations, and calculate the true response value of each point, that is\left \{x^{n+1},...,x^{N+q} \right \},\left \{y(x^{n+1}),...,y(x^{n+q}) \right \}

KB:

Remember the proxy model established by the current known data\tilde{Y}(x)

Select q points in each iteration, and each point is no longer the real response value, but the value of the proxy model, that is,\left \{x^{n+1},...,x^{N+q} \right \},\left \{\tilde{Y}(x^{n+1}),...,\tilde{Y}(x^{n+q}) \right \}

 

CL:

Remember the known points Y, the current constant is L, where L can be min(Y), mean(Y),max(Y)

Select q points in each iteration, and each point is no longer the real response value, but the value of the proxy model, that is,

 \left \{x^{n+1},...,x^{N+q} \right \},\left \{L,...L \right \}

 

Guess you like

Origin blog.csdn.net/qq_44785318/article/details/123644298